Local Traffic Evironment Perception For Vehicle Based On Multi-Sensor

Smart vehicle is the ultimate form of the future vehicle development, making use ofperceptual system instead of the driver’s perception of traffic environment. Trafficenvironment perception systems of vehicle use different sensors to detect vehicles andsurrounding environment, providing the basis for the entire smart vehicle driving strategyand path planning. In the part of perception of intelligent vehicle, it mainly relies on thecoordinate operation of vision system and radar system. The majority of domestic researchesare the use of visual system more than of radar system; however, the foreign researchinstitutions have done a lot of work. Taking advantage of single-sensor has following flawscomparing with the multi-sensor system: Low confidence of measurement and dimension ofsystem; Poor system reliability and unstable performance; Low utilization of systemresources.According to the limitation of single-sensor system in perceiving traffic environment,this paper has done the research on perception of vehicle traffic environment on the basis oflessons from both foreign and domestic perceptive technology of intelligent vehicle.The multi-sensor perception system of traffic environment includes millimeter-waveradar, laser and vision system. Vision system is used for lane detection and tracing movingcar; the millimeter-wave radar can provide relative speed, range and offset angle of obstacledetected; the laser can acquire the contour information at the plane it scans. This paper hasalso established the vehicle platform equipped with multi sensors, and developed theintelligent vehicle software platform, and study the sensing algorithm based on this platform.In the millimeter-wave radar and laser radar systems, the researches on receivingmillimeter-wave radar data and data filtering are conducted, as well as the data receiving oflaser data through serial communication technology. Then installed the radar system on the test vehicle and completed the unified calibration of sensors and coordination unification,finally, accomplished the perception of traffic environment on the base of radar system.This paper used vision system to identify the road lane. Compares and analyzes thecurrent image processing technology for lane detection method, considering reliability andreal-time performance, then selects a high reliability and good real-time lane detectionalgorithm, that is, use progressive threshold segment method for image binarization, whichcan resist interference from noise, combine the Hough transform to recognize marking linesaccurately. Making use of the previous image of last frame to set region of interest canreduce the time cost of Hough transform and make the lane recognition more accurate.Finally, accomplish the perception of lane marking line on the base of vision system.The vehicle detection is another important role in the visual system. In this paper, theshadow of the bottom of the vehicle is the region of interest, extract the contour of binaryimage with canny operator, integrate or remove the region of interest by rectangle constraintdiscipline, then calculate entropy of the region of interest to judge the texture complexity,and finally using the Robinson operator to extract contours of the vehicle. In a continuousimage sequence, the use of classical Kalman filter to predict the detected obstacle isproposed, which can improve the stability of the vehicle detection algorithm and reduce thecomputation time. The tests showed that the algorithm can track vehicles stably, even whenthe objective car is changing lane.In the test section, this paper studied calibration for both monocular camera andbinocular camera system, obtaining the camera’s internal and external parameter.Furthermore, make use of disparity map of binocular vision system to measure the distanceof obstacles, which can obtain the position in the camera coordination system. Aftertransforming the camera coordination into vehicle coordination, the position relevant tovehicle can be obtained. Finally, the paper establishes the model of transformation betweenthe camera coordinate system, radar coordinate system and vehicle coordinate system. Selectany two points on the marking line detected in the image, the lane line can be projected tothe coordinate system of the vehicle. Lane line information and radar detection results will eventually be expressed in the form of a bird’s eye view，which can provide a basis for futuremulti-information fusion research and vehicle driving Strategies.At length, perception results of the various parts of sensor are expressed in one samebird eye view, achieving the description of traffic environment.